> ## Content Index
> Fetch the complete content index at: https://f4n6.co.uk/llms.txt
> Use this file to discover other available public pages before exploring further.

# COLD RELIC leverages cloud data theft for cross-organization impersonation
- URL: https://f4n6.co.uk/security-feed/cold-relic-leverages-cloud-data-theft-for-cross-organization-impersonation/
- Published: 2026-11-17T20:24:14.000Z
- Updated: 2026-11-17T20:24:14.000Z
- Author: Jeff Davies
- Tags: #security-feed, Cold Relic, ColdRiver, UNC4057, Star Blizzard, Callisto

## 1\. Executive summary

Mandiant has published an incident response case study of two linked intrusions by the Russian state-sponsored group COLD RELIC, which the US Department of Justice has assessed operates within Russia's Federal Security Service (FSB). In a late-2025 intrusion at a US organisation the group used stolen Super Admin credentials, an MFA push-fatigue attack and a US-geolocated residential proxy to enter a Google Workspace tenant, then mass-shared and downloaded the organisation's cloud storage, exported the full staff directory and ran a tenant-wide Google Workspace Customer Takeout. In a 2026 intrusion at a second US organisation the group impersonated an employee of the first victim over an encrypted messaging app, sent phishing links that harvested Microsoft Entra ID credentials and tokens, and registered a new MFA device to hold the account. No CVE is involved and no EMEA financial services victim is named; the risk to clients is identity and cloud-control weakness that lets a single compromised account become tenant-wide data theft and a pretext for attacking partner organisations.

## 2\. Regulatory framing

| Article                                                 | Trigger (the fact in this item)                                                                                                                                                                         | Practical impact                                                                                                                                                        |
| ------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| DORA Art. 17: ICT-related incident management process   | The victim in the first case reset passwords but did not revoke existing sessions, and COLD RELIC kept access for several months; the actor also deleted Alert Center notifications to remove evidence. | The incident process must require full session, refresh-token and OAuth revocation plus preservation of cloud audit logs, not password reset alone.                     |
| DORA Art. 28: ICT third-party risk — general principles | The second intrusion entered through Microsoft Entra ID used as the federated SSO provider for Google Workspace, and the actor moved between cloud ecosystems.                                          | Third-party and cross-provider identity dependencies (federated IdP, cloud productivity tenant) need to be in scope of provider oversight and exit/continuity planning. |
| NIS2 Art. 21(2)(d): supply chain security measures      | COLD RELIC used contact and relationship context from the first victim to phish the second, and repurposed a compromised account into a new persona to target the same professional circles.            | Supply chain security must cover the identity and communications exposure created by partners and counterparties, not only software suppliers.                          |

## 3\. Technical analysis & attack chain

The two cases are linked: phishing URL parameters and impersonation pretexts in the 2026 intrusion indicated that COLD RELIC likely used contact and relationship context from the late-2025 victim to target the second organisation. Mandiant states this as a likelihood, not a confirmed fact. GTIG observed evidence of new COLD RELIC phishing operations as recently as November 2026.

Confirmed steps in the late-2025 Google Workspace intrusion:

1. COLD RELIC routed access through a residential proxy network in the US located in close geographical proximity to the victim, avoiding impossible-travel and similar heuristic alerts.
2. The actor authenticated with stolen credentials of a Super Admin account and ran an MFA push notification fatigue attack. After the victim user accepted an MFA request, COLD RELIC registered a device model previously unseen in the tenant.
3. The actor used passkeys, MFA backup codes and rogue MFA devices to retain access across the intrusion.
4. With Super Admin privileges, COLD RELIC downloaded the contents of the organisation's entire cloud storage drive, both by mass sharing files to a COLD RELIC-operated email account and by downloading files directly.
5. The actor accessed the Admin Console, queried Admin security investigation and email audit logs to check defensive visibility, and viewed, modified and deleted Alert Center notifications to remove evidence of the initial intrusion.
6. COLD RELIC exported full employee lists and initiated a domain-wide data export via Google Workspace Customer Takeout. Customer Takeout is an administrative tenant-wide export tool, distinct from end-user Google Takeout; requests are logged in Google Workspace Admin audit logs as CUSTOMER\_TAKEOUT\_CREATED, and administrators can disable Google Takeout for users at the organisational level.
7. The actor kept access for several months and performed email inbox sync via OAuth to external email clients, reauthenticated with unrevoked attacker-enrolled passkeys that persisted across password and session resets and bypassed device registration controls, and made targeted cloud drive document access for credentials, email archives, personnel information and other documents matching its collection requirements.
8. After COLD RELIC was removed from the infrastructure, GTIG observed email accounts impersonating employees of the victim organisation being used against internal employees and associates in the same professional circles.

Confirmed steps in the 2026 Microsoft Entra ID intrusion:

1. COLD RELIC impersonated an employee of the first victim organisation to open communications over a secure messaging app, then shared phishing links during the conversation.
2. The links were designed to steal Entra ID credentials and authentication tokens and redirected the victim to document lures relevant to the conversation to avoid suspicion.
3. After successful authentication with the stolen credentials, COLD RELIC registered and used a new MFA device in the SSO solution to retain access.
4. The actor obtained credentials to the organisation and performed a bulk download of files accessible in the compromised user's cloud drive.
5. After the compromise, the encrypted messaging application recorded the actor changing their profile name to a completely new person. GTIG believes this persona was likely used to target other organisations in the same professional circles.

Phishing infrastructure. COLD RELIC typically uses a compromised website as the initial link that redirects the target to the actual phishing domain, which hides the phishing domain and lets the actor fingerprint visitors to decide whether to redirect them. Links usually carry a parameter whose value the attacker sets, often indicating a relationship to a previous victim or the lure topic. Mandiant's examples are `https://compromised.example.com/some_path/some_file.php?file=<previous_victim_name>` and `https://compromised.example.com/some_path/some_file.php?ref=<lure_name>`, and it notes this is not always the case. In the 2026 case, GTIG correlated a COLD RELIC redirector URL submitted to VirusTotal in June 2026 whose query parameter referenced the first victim organisation.

Collection priorities. Post-compromise, the group's primary objective is mass collection and download of files and email archives, alongside targeted queries for credentials, personnel information, email archives (.pst files) and documents matching its collection requirements. It also seeks organisational rosters, contact lists, resumes and event schedules to build pretexts for later targets. Mandiant notes that in some cases users stored credentials and other secrets in cloud storage, which let COLD RELIC pivot to additional accounts and services including personal accounts unmonitored by employers.

Attribution. Mandiant names COLD RELIC and links the former tracking names COLDRIVER and UNC4057 and the aliases Star Blizzard, Callisto and Gossamer Bear. The DOJ assessment that the group operates within the FSB is reported by Mandiant. MITRE ATT&CK holds a profile for Star Blizzard (G1033); it holds no profile for COLD RELIC, UNC4057 or Gossamer Bear, so the equivalence of these names rests on Mandiant's reporting and should be treated as unconfirmed for tooling that keys on ATT&CK group IDs.

## 4\. Mitigation & containment

P1 within 24 hours

- Revoke all active browser sessions, refresh tokens and third-party OAuth app permissions for any account suspected of compromise, across both Google Workspace and Microsoft Entra ID, and remove unauthorised passkeys, backup codes and MFA devices. Mandiant attributes at least one case of continued access to a password reset that left sessions live.
- Replace SMS, voice and push MFA with phishing-resistant FIDO2/WebAuthn hardware security keys (Mandiant names Titan Security Keys). Hardware keys are bound to the origin URL and cannot be intercepted by adversary-in-the-middle reverse proxies or approved through fatigue attacks.
- Alert on Google Workspace Admin audit events CUSTOMER\_TAKEOUT\_CREATED, User List Download, Investigation Query, Batch Delete Operation Performed, Passkey enrolled, Device Registered, and Drive User Sharing Permissions Change or visibility changed to shared\_externally. Treat any tenant-wide export or Alert Center deletion as an incident.
- Restrict administrative logons from consumer ISPs, VPNs and known residential proxy ASN ranges; Mandiant names AS5650 and AS20115 as examples.

P2 within 72 hours

- Enrich authentication logs with residential proxy and VPN IP reputation feeds and alert on sudden residential ASN shifts for privileged accounts. Apply conditional access policies to block or detect logons from known proxy networks.
- Require existing administrators or out-of-band identity verification to approve new MFA devices or secondary authenticators such as Microsoft Authenticator or passkeys.
- Review authorised third-party applications in Google Workspace and Microsoft 365 and flag unexpected native desktop email clients (Mandiant names Thunderbird) or external mail services (Mandiant names Proton Mail). Alert on new OAuth authorisations granting broad scopes such as `https://mail.google.com/` to unapproved clients.
- Deploy DLP rules across Google Drive, OneDrive and SharePoint to alert on or block documents containing sensitive terms such as passwords, roster or .pst, and to restrict mass external file-sharing. Google Workspace Security & Data Protection is named as one option.
- Consider enrolling privileged users in the Advanced Protection Program.

P3 within 7 days

- Add session and OAuth revocation to BEC playbooks as a mandatory step alongside password reset.
- Audit where credentials and secrets are stored in cloud storage and remove them, given the pivots Mandiant observed.
- Review detection coverage against the Google SecOps rule names Mandiant lists: Google Workspace Successful MFA Fatigue Attack, Google Workspace Customer Takeout Initiated, Google Workspace Alert Center Alert Deletion, Google Workspace Directory User List Download, Google Workspace Drive Bulk External or Freemail File Sharing, Google Workspace Device or Passkey Enrollment Following MFA Fatigue, and Google Workspace OAuth Broad Mail Scope Granted to Unapproved Client.

No CVE, patch or fixed version applies to this item; the mitigations are configuration and identity controls.

## 5\. Indicators of compromise

No atomic indicators (hashes, domains, IP addresses, file paths) are published in the source. The URL patterns Mandiant gives use the placeholder domain compromised.example.com and are illustrative, not live infrastructure. The behaviours below are the pivotable observables.

| type      | value                            | confidence | source                       |
| --------- | -------------------------------- | ---------- | ---------------------------- |
| behaviour | see behavioural indicators table | n/a        | Google Cloud Blog / Mandiant |

Behavioural indicators

| behaviour                                                                                                                        | where to observe                                                                                   | confidence      |
| -------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------------- | --------------- |
| Multiple failed logins or MFA challenges for one user in a short window followed by a successful authentication from the same IP | Identity provider authentication logs (Google Workspace, Entra ID)                                 | High (Mandiant) |
| New OAuth authorisation granting broad mail scope to a non-standard client such as Thunderbird or Proton Mail                    | Google Workspace OAuth logs, Authorize event                                                       | High (Mandiant) |
| Tenant-wide export created                                                                                                       | Google Workspace Admin audit log, CUSTOMER\_TAKEOUT\_CREATED                                       | High (Mandiant) |
| Spike in file downloads by one user from non-corporate ASN or residential proxy                                                  | Google Workspace Drive logs, Download events                                                       | High (Mandiant) |
| Spike in documents shared externally, particularly to free email providers such as @gmail.com                                    | Google Workspace Drive logs, sharing permission change or visibility changed to shared\_externally | High (Mandiant) |
| New passkey or device registration, especially after an MFA fatigue pattern or from unexpected geography                         | Google Workspace Passkey enrolled, Device Registered events                                        | High (Mandiant) |
| Authentication using an MFA backup code from a non-corporate ASN or proxy network                                                | Google Workspace User logs, Login Challenge Type = Backup code                                     | High (Mandiant) |
| Security alerts modified or deleted, including Government-backed attack alerts                                                   | Google Workspace Admin logs, Batch Delete Operation Performed                                      | High (Mandiant) |
| Domain user list export                                                                                                          | Google Workspace Admin logs, User List Download                                                    | High (Mandiant) |
| Audit or security log queries by an unexpected user                                                                              | Google Workspace Admin logs, Investigation Query                                                   | High (Mandiant) |
| Drive keyword searches for .pst, .bak, password or cred                                                                          | Google Workspace Drive logs, Item search performed                                                 | High (Mandiant) |
| Logon from residential proxy ASN geolocated near the victim                                                                      | Authentication logs enriched with proxy/VPN reputation                                             | High (Mandiant) |

## 6\. Detection

The source gives log event names and behavioural patterns rather than file artefacts, so a Sigma rule is the appropriate form. No YARA rule is written because the source contains no file strings, hashes or command lines from COLD RELIC tooling.

```yaml
title: COLD RELIC Cloud Identity and Data Theft Behaviours
id: 8f2c1a44-cold-relic-gws
status: experimental
description: Detects Google Workspace and Entra ID activity consistent with COLD RELIC post-compromise behaviour as described by Mandiant.
author: Adverse Trace
date: 2026/11/17
references:

  - https://cloud.google.com/blog/topics/threat-intelligence/cold-relic-a-case-study-in-impersonation-and-compromise/
logsource:
  product: google_workspace
  service: admin
detection:
  takeout:
    event.name: 'CUSTOMER_TAKEOUT_CREATED'
  alert_tamper:
    event.name: 'Batch Delete Operation Performed'
  directory_export:
    event.name: 'User List Download'
  recon_query:
    event.name: 'Investigation Query'
  device_enroll:
    event.name|contains:

      - 'Passkey enrolled'
      - 'Device Registered'
  backup_code:
    event.name: 'login'
    Login Challenge Type: 'Backup code'
  sensitive_search:
    event.name: 'Item search performed'
    search_term|contains:

      - '.pst'
      - '.bak'
      - 'password'
      - 'cred'
  condition: takeout or alert_tamper or directory_export or recon_query or device_enroll or backup_code or sensitive_search
falsepositives:

  - Legitimate administrative exports and audits; correlate with source ASN and user role
level: high

```

For MFA fatigue, alert on repeated failed or MFA-challenge events for one user followed by a success from the same IP. For OAuth abuse, alert on new authorisations granting `https://mail.google.com/` to unapproved clients. Field names above follow the event names Mandiant lists; map them to your own log schema before deployment.

## Threat actor context

**Star Blizzard** · [G1033](https://attack.mitre.org/groups/G1033?ref=f4n6.co.uk) · aka SEABORGIUM, Callisto Group, TA446, COLDRIVER

[Star Blizzard](https://attack.mitre.org/groups/G1033?ref=f4n6.co.uk) is a cyber espionage and influence group originating in Russia that has been active since at least 2019\. [Star Blizzard](https://attack.mitre.org/groups/G1033?ref=f4n6.co.uk) campaigns align closely with Russian state interests and have included persistent phishing and credential theft against academic, defense, government, NGO, and think tank organizations in NATO countries, particularly the US and the UK.

*No MITRE ATT&CK profile for: COLD RELIC, UNC4057, Gossamer Bear.*

## 7\. Sources

- Google Cloud Blog / Mandiant, "COLD RELIC: A Case Study in Impersonation and Compromise", 2026-11-17\. https://cloud.google.com/blog/topics/threat-intelligence/cold-relic-a-case-study-in-impersonation-and-compromise/
- MITRE ATT&CK, Star Blizzard (G1033). https://attack.mitre.org/groups/G1033/

## 8\. Adverse Trace position

Adverse Trace assesses this as a high-impact identity and cloud data-theft campaign with no CVE and no named EMEA financial services victim, so direct exposure to our client base is not established by the source. The material risk is control weakness: a Super Admin account protected only by push MFA, sessions left live after a password reset, and credentials stored in cloud drives are each sufficient for tenant-wide theft and for the actor to reuse the victim's identity against partners. The attribution to Russian FSB-linked espionage rests on Mandiant's reporting and the DOJ assessment it cites; MITRE ATT&CK holds a profile only for Star Blizzard, so the COLD RELIC, UNC4057 and Gossamer Bear names are unconfirmed against ATT&CK. We will track GTIG reporting on new COLD RELIC phishing through November 2026 and later, and we recommend clients prioritise phishing-resistant MFA and full session and OAuth revocation in their BEC playbooks this week.

---

[Read the original source →](https://cloud.google.com/blog/topics/threat-intelligence/cold-relic-a-case-study-in-impersonation-and-compromise/?ref=f4n6.co.uk)

*Published via PulseTrace — Adverse Trace threat intelligence.*